Probability Plot Result Comparison with Recurrent Neural Network Approach for Path Navigation of a Humanoid in Complex Terrain

This research work utilizes the concept of recurrent strategy of neurons, which performs sequential tasks where the output and input data are dependent with each other. The major advantage of using recurrent neural network (RNN) for humanoid motion planning lies in spending the previous used long sequence information through memory. RNNs form direct cycles having internal state and form prime candidate for handling learning procedure. In this paper, long short-term memory (LSTM) RNN is implemented in humanoid robot to test the motion planning analysis. In the neural network model, the obstacle distances from robot’s location are fed as input parameters, and moving angle (MA) is obtained as the output parameter from RNN to guide the humanoid to reach the target with LSTM. Both simulation and experimental navigations are carried out through the developed technique. Probability plot between the simulation and experimental results is performed with normal distribution with comparison analysis. It is found that the results are satisfactory for humanoid navigation. The percentage of deviation between simulation and experimental results in terms of navigation variable is below 6%, which is in acceptable limit range.